Sie sind auf Seite 1von 5

Journal of Ship Production, Vol. 23, No. 4, November 2007, pp.

197–201

Shipyard Space Allocation and Scheduling

Daniel A. Finke,* Christopher B. Ligetti,* Mark T. Traband,* and Allan Roy†


* Applied Research Laboratory, Pennsylvania State University, State College, Pennsylvania

Northrop Grumman Newport News, Newport News, Virginia

In this paper we describe the development of a tool that allows shipyard planners to
efficiently and effectively plan space within valuable areas of a shipyard. Traditionally,
space is considered a resource; however, it is difficult to accurately account for and
plan its consumption with the current planning software tools available. The spatial
scheduling tool described in this paper can be used by planners and construction
management teams to manually or semiautomatically reserve space within the ship-
yard for construction of large units over the entire ship erection period. The result is
the ability to efficiently generate and compare multiple space allocation plans with the
ultimate goal of maintaining the critical ship erection schedule. This work has been
developed with Northrop Grumman Newport News (NGNN) under a project funded by
an Office of Naval Research (ONR) Manufacturing Technology (ManTech) program.

1. Introduction blies, and hundreds of thousands of parts. While the aircraft car-
rier’s size dictates more units than the DDG, significantly higher
SHIPBUILDING HAS significantly evolved throughout the past lifting capabilities allow the carrier to be constructed using much
quarter century. The traditional “stick-building” approach to ship- larger units than the base units used to construct the DDG.
building has been replaced with a modular construction strategy as As the units become larger and heavier, production space in the
dry-dock and lifting capabilities at shipyards have improved. This shipyard becomes a constraint. The larger units are limited in
change enabled shipbuilders to effectively outfit larger sections of where they can be produced due to the lifting and handling limits.
the ship earlier in the construction process rather than relying on For this reason, it is important to accurately plan the space in these
less efficient practices that characterized dry-dock construction. areas to ensure that units are moved only when and where neces-
Using a modular construction strategy, relatively small piece parts sary to efficiently use the available space. Unnecessary moves
are joined to form subassemblies, subassemblies are joined to result in nonvalue-added cost to the unit. The maximum number of
form assemblies, units or blocks, and units are joined together to units should be constructed in the high-value areas of the shipyard.
form larger blocks of the ship. Typically, the ship is then erected However, due to production constraints and aggressive construc-
block-by-block until the ship is complete. The various units (sub- tion schedules, maximizing the number of units in an area may
assemblies, assemblies, units, and blocks) are almost always result in unnecessary moves, while minimizing unnecessary
unique in size, shape, and weight and largely depend on the size moves results in a less efficient use of the space.
of the final ship, the lifting and handling capabilities of the spe-
Spatial scheduling is currently being done by small groups of
cific shipyard, and the extent of outfitting that is completed in that
experienced people using tools such as computer-aided design
area of the ship.
(CAD), PowerPoint, or Excel and schedule information from their
For instance, construction of a CVN aircraft carrier hull struc-
planning systems. Although these ad hoc tools are relatively ef-
ture is made up of approximately 170 super-sized units, more than
fective, they are cumbersome and require a significant amount of
300 smaller erectables, and 1,500 base units, in addition to the
time to update even for minor schedule changes. In addition, spa-
assembly of thousands of details and millions of parts. On the
tial scheduling practices and any lessons learned over time are
other hand, the construction of a DDG, which is obviously smaller
contained within the experts themselves. Over time, this knowl-
than CVN, requires roughly 200 base units, hundreds of assem-
edge is lost and must be reacquired by yet another generation of
new employees. Providing some type of modeling or algorithmic
support to capture this knowledge and automate the process with
Manuscript received by JSP Committee from SPS 2006. a “smarter tool” would provide a more efficient allocation of the

NOVEMBER 2007 8756/1417/07/2304-0197$00.00/0 JOURNAL OF SHIP PRODUCTION 197


valuable production space in each of the construction areas/
facilities.

2. Problem

The dynamic allocation of space in the shipyard is an im-


mensely difficult and time-consuming effort. The difficulty in
scheduling floor space, or spatial scheduling, arises in the fact that
the allocation of space to one block significantly affects the avail-
ability of floor space to every other block. Scheduling production
space to satisfy an erection schedule becomes even more complex Fig. 1 Bin-Packing problem
when unexpected changes to the schedule occur (e.g., upstream
process delays, weather-related delays, or subcontractor timeli- the general case of the academic problem is not relevant in the
ness). This illustrates a need for a tool that can assist planners in practical sense due to the fact that the “real-world” system has
not only generating efficient spatial layouts, but also modifying significantly more complex constraints than those of the general
these plans accordingly with minimal additional effort. case. Some of these constraints include preferred locations, spac-
Not only is the practice of scheduling space a difficult problem, ing between the units, schedule requirements, and so forth. The
but the automatic or semiautomatic scheduling of the space is even general formulation is difficult to solve even without the added
more difficult. In academia, this problem follows three lines of complexity of the practical constraints.
research: scheduling, cutting stock, and bin packing. The bin-packing problem most closely represents the spatial-
Scheduling problems are usually characterized by allocating planning problem that faces planners in the U.S. shipbuilding
work to fixed resources that have known capacity in a single industry. Although there is little research on this problem, the
dimension (time). Traditionally, problems such as this have been research that has been done has been in the shipbuilding industry
transformed into single-dimension problems by creating uniform and is applicable to this problem. The next section describes the
work areas and treating those areas as fixed resources. Of course, approach that we have developed to help in the allocation and
by truncating each unit to a specific work area size, a significant planning of floor space within the shipyard.
amount of fidelity is lost in the transformation process and thus
results in a less efficient layout and the need for conservative 3. Approach
allocations.
The cutting stock problem is a well understood problem in the The objective of the tool described in this paper is to increase
shipbuilding industry. Steel processing facilities in almost every the utilization of work area space, while maintaining production
shipyard use nesting software to determine the best allocation of schedules. In addition, the ability to make changes quickly and to
steel plate area for cutting out profiles. This problem extends the view the impact of those changes in real time provides a powerful
basic scheduling problem into a two-dimensional solution ap- tool that will significantly reduce the cost of planning and replan-
proach (length and width). By extending the problem from one to ning.
two dimensions, the allocation of steel plate space to the profiles The current method of spatial scheduling is largely a manual
is much more efficient and results in less waste. Solution proce- process with the use of software packages that make the process
dures to the two-dimensional cutting stock problem have and con- cumbersome. However, there is a wealth of information that has
tinue to be developed to improve the efficiency and computation been developed and retained by the planners in an effort to un-
time of the plate layout. derstand and plan the layout of the work areas within any ship-
The theoretical definition of the “bin-packing problem” is simi- yard. This expert knowledge only resides in the planners them-
lar to the dynamic space allocation problem that we must solve to selves and is key to efficient planning. Some planners have been
automate or semiautomate the allocation of space within a ship- scheduling floor space and collecting the information to do so for
yard. In this problem, cubes or solid boxes are “packed” into a decades.
larger empty container in an effort to maximize the number of An incremental development approach has been taken in this
boxes in the container. In the shipbuilding context, the work area, research. We knew that a tool that simplified the planner’s current
platen, or shop floor length and width are considered the “X” and tasks would improve the process, but at the same time the method
“Y” dimensions of the container and the “Z” dimension is the by which planning was done did not change drastically. In addi-
schedule horizon. Figure 1 shows a visual representation of the tion, the vast amount of knowledge residing with the planner could
problem. Only a few solution procedures have been developed for be used implicitly to develop plans. In the first phase of the tool
these types of problems (Lee et al. 1996; Lee et al. 1997; Lodi et development, a graphical user interface was developed to help
al. 2002; Lim et al. 2005), and optimal solutions procedures have with the planning task.
proved to be intractable. The semiautomated procedure described The second phase of the tool development focused on capturing
in this work is similar to that of Lee et al. 1996. the knowledge of the planners and using it to develop an auto-
There is one key difference between shipyard spatial scheduling mated or semiautomated procedure to allocate space within the
and the conventional “bin-packing” problem. In the bin-packing shipyard. A procedure such as this is often called an expert system.
problem, it is generally assumed that the blocks to be packed are The system would use the knowledge of the planners to help make
all available at time 0. In shipbuilding, the blocks become avail- similar decisions in placements, but also use the power of the
able for placement at different times, resulting in potential “float- computer to provide a “look-ahead” capability that could antici-
ing blocks” in the “bin-packing” context illustrated in Fig. 1. Also, pate spatial conflicts in the future.

198 NOVEMBER 2007 JOURNAL OF SHIP PRODUCTION


The following sections provide further details on the two phases Additional features included in the spatial scheduling tool in-
of the tool development: the graphical user interface and the au- clude:
tomated or semiautomated scheduling procedure.
• Space utilization over time graph
• Floor plan printing
3.1. Graphical user interface • Saving/loading of spatial layouts
The first part of the tool is an interface for the user (usually a • Zooming for higher or lower level of granularity
planner or construction manager) to interact with the unit at- • Measuring distance between units.
tributes, schedule information, and the actual placement of the While the spatial scheduling tool provides a planner with several
units within a shop or production work area. The graphical user features to generate efficient spatial plans more rapidly, the actual
interface (GUI) developed for this tool allows the user to select a method of allocating space is not much different than current
unit from a list that has been populated with known or expected shipyard practices, where block placement decisions are based on
unit attributes such as length, width, height, weight, and so forth expert-user knowledge. The following section describes a method
and schedule information such as scheduled start date, planned to automatically allocate space according to this expert knowl-
duration, early start, late finish, actual start, and production com- edge. While it is nearly impossible to capture the entire set of
plete information. Figure 2 shows the GUI annotated with the key rules, constraints, and preferences used to generate a near-optimal
features. spatial layout, the automated scheduler can be used to generate a
The user would begin using the tool by searching for a sched- valid baseline layout, and the end-user can make modifications to
uled block placement in the schedule window. Scheduled place- this layout using the spatial scheduling tool GUI.
ments are colored to assist the user in determining which blocks
need to be placed: black to identify a block scheduled to be placed
3.2. Automated/semiautomated scheduler
at the current time period, gray to identify a block scheduled to be
placed in a future time period, and red to identify a block sched- There are several techniques available to help solve this type of
uled to be placed in a previous time period. When double-clicked, problem. Over the course of this project we have investigated two
the block appears on the plan view and can be moved to any techniques: mathematical programming and expert systems.
location using a standard PC mouse or keyboard. Attributes to any Investigations into the mathematical programming solutions re-
block can be viewed by selecting it in the schedule window or the sulted in a nonlinear, integer program, where the objective was to
plan view. place all scheduled units such that the total difference between the
The user can move forward or backward in time to view the actual placement time and the scheduled placement time was mini-
spatial layout for any period during the planning horizon. As they mized. The constraints to this model included no early placement
do so, it is possible that certain blocks will appear or disappear, of units, units remained on floor for entire planned duration, and
depending on when those blocks were placed and when they are no units could intersect with any other unit for its entire construc-
scheduled to be complete. The user is also notified of any colli- tion duration. Solutions to this mathematical programming model
sions between overlapping blocks with a light on the GUI. If this proved to be infeasible in terms of the required processing time for
light turns red, it is indicated to the user that two or more blocks a solution. This result was obtained by testing the procedure on
are overlapping; otherwise, the light will be green. The collision several small problems, where the run time exceeded a reasonable
check does not only occur for the current time period, but instead time limit. The solution time would only increase more when
checks for collisions over the entire planning horizon when a new considering additional rules and preferences identified by shipyard
block is placed, a block is moved or resized, or when a block’s experts. For this reason, alternative solution strategies have been
schedule is modified. By clicking on the red light, the user is sought out, specifically expert systems.
informed which blocks are overlapping and when in time the An expert system is a problem-solving technique that uses ex-
collision is occurring. pert knowledge (that of the planner or construction manager) to
guide a computer-based search to solve complex problems such as
the one described here.
The methods in current literature can be divided into two cat-
egories: online and holistic. Online methods try to allocate floor
space to a unit as the unit becomes available to schedule. A ho-
listic approach is not limited to scheduling the first unit on the list;
it can schedule any unit at any time period in an effort to find the
most efficient layout. Online scheduling occurs in systems where
there is no set plan or schedule for units or the system is so
dynamic that there is no way of knowing the “state” or layout at
any given time. Holistic scheduling problems can be much more
difficult to solve than online ones, because there is much more
information about the state of the system that needs to be taken
into account.
The method that we have developed is a holistic approach that
uses the strengths of an expert system. The procedure is further
defined as a rule-based expert system, in which expert and em-
Fig. 2 Spatial scheduling tool graphical user interface pirical knowledge have been condensed into a set of rules. These

NOVEMBER 2007 JOURNAL OF SHIP PRODUCTION 199


rules are then used to determine the placement of the units within
the allowable floor space. Both the graphical user interface and the
rule-based system were developed in Java, and solution times are
reported for trials performed on a Dell Precision M60 with a 1.7
MHz processor.
The conceptual model for this problem is a three-dimensional
grid, where the length and width are the length and width of the
work area, and the height is the planning time horizon. Similar to
the GUI, the planning horizon can be thought of as a finite time Fig. 3 Conceptual model of space allocation problem
period, that is, 5 years for the construction of an aircraft carrier, or
a rolling “window” of time, that is, 3 years from the current date. The system takes the production preferences and constraints
Figure 3 shows the conceptual model used to solve this problem. into account when trying to place the units in the work space. For
The figure illustrates how a block consumes space for each time example if production supervisors or planners prefer to have a unit
period from the period it is placed to the period it is completed and placed near a doorway, then the algorithm tries to place the unit
leaves the work area. next to the doorway. The algorithm is designed to run iteratively
The dimensions for the grid cells need to be defined to a level so that several placements of all of the blocks will be generated.
of detail that is dictated by the scheduling precision. In other The various layouts that are generated will have a certain number
words, if the planners in the shop are planning to the square ft. of preferences that have been met. The layouts that have more
then the length and width of each cell should be 1 ft; however, if preferences met would be more desirable.
a placement of a unit is required to be to the inch then the cell
should be to the inch. The height of each of the cells is defined by 3.3. Algorithm
the planning period increment (e.g., weeks, days, etc.).
It is important to note that the smaller the cells in the grid are The following algorithm outlines the steps used to solve this
the more time the procedure described in this paper and any other type of problem to get a single baseline layout that a user could
procedure would take to solve this problem. In some instances, it take and manipulate.
may be better to sacrifice precision in an effort to decrease the Step 1: Initialize the system, by reading the schedule, calculate
solution time of the procedure. Also, the level of detail should not the CP for each period and create the period list.
be finer than the level of detail that can be obtained and main- Step 2: Pull and remove the first period from the period list and
tained on the shop floor. For instance, if the units cannot be placed set it to the constraint period. If the list is empty go to
with accuracy to the inch then the tool should not be planning to Step 4.
the inch because it would take a long time to solve, and the Step 3: Use rules and heuristic search to position units on floor
solution that would be generated could not be guaranteed on the space. If all units are placed then go to Step 4. If cannot
shop floor. place all units, then halt search and alert the user. Else
Through initial empirical analyses of the problem we found that go to Step 2.
by placing units in periods that were congested first, it was easier Step 4: Record the number of preferences that have been met
to allocate floor space in the periods prior to and after the con- and present solution to the user.
gested period. We also found that the congested period was not
necessarily the period that had the most units allocated to it or the This represents the semiautomated approach where a significant
one with the highest utilization, but instead the period that had the amount of planning and analysis would still be required by the
most square footage assigned to it with the minimum number of planner to ensure that all of the constraints of the shipyard were
units. Equation (1) was used to calculate a constraint period (CP) met. A more completely automated approach to this problem
value for each planning period. would be to iterate on the above algorithm in an effort to maximize
the number of preferences that have been met.
CPi = SFi Ⲑ SF range − Ni Ⲑ NR
i = 1, 2, 3, . . . , P, (1) 4. Results/conclusions

where: For small problems (∼20 units) solution times are between 9 and
300 seconds. For much larger problems (>100 units), the time
P is the total number of time periods in the schedule. required to find a single solution increases to between 4 minutes
SFi is the required raw square feet for period i, and several hours. In comparison, preliminary tests on the math-
SF range is the maximum required raw square feet minus the ematical programming procedures showed that many smaller
minimum for the entire planning horizon. problems (∼3 to 4 units) were unsolvable using current optimiza-
Ni is the number of units scheduled in period i. tion software.
NR is the maximum number of scheduled units minus the mini- In some cases, there may not be a feasible solution to the layout
mum for the entire planning horizon. problem. When there is no feasible solution, the procedure halts
after a predetermined time and reports to the user that there are too
Each of the time periods in the planning horizon is given a value many units that need to be allocated to the space. This procedure
according to equation (1) and inserted into a list in nondecreasing has been developed from a class of solution procedures called
order of CP. The procedure then moves down the list and places heuristics. Unlike mathematical programming, heuristics are not
each of the units. guaranteed to find an optimal solution or any feasible solution for

200 NOVEMBER 2007 JOURNAL OF SHIP PRODUCTION


that matter. It is necessary to use these types of procedures be- especially when considering more sophisticated rules and con-
cause of the complexity of the problem under consideration. straints. In addition, methods of gathering rules, constraints, and
In the first incremental phase of the tool development, we found preferences from expert users (i.e., rule elicitation) and translating
that there was no significant time savings in the overall planning these to computer code also warrant investigation. This would
effort of the floor space. The time savings that we expected was greatly facilitate the implementation in other areas of the shipyard
used to do “what-ifs” and develop alternative and contingency or other shipyards entirely where different rules and constraints
plans in an effort to generate the most efficient space plan. The are present. In addition, this would be a vehicle for knowledge
automated procedure shows merit as a tool to assist planners and capture and retention for future planners to use.
shop supervisors in the layout and planning of space in the ship- Finally, there are often height constraints in shipyard construc-
yard. tion facilities and platens, usually due to the crane hook height. An
additional level of complexity is introduced when there are height
constraints, for the spatial scheduling problem evolves into a four-
5. Future work
dimensional problem (length, width, schedule, and height). Future
The work outlined in this paper presents several areas for future automated placement routines should consider how the inclusion
research and development for shipyard spatial scheduling. Nearly of the height constraints impacts the development of an automated
all shipyards experience similar problems when it comes to sched- space allocation routine.
uling space. Given the lack of commercial off-the-shelf (COTS)
software for scheduling space over time, a yard-configurable tool References
similar to the graphical user interface (GUI) previously discussed
could allow for quick expansion to other shipyards by allowing LEE, K. J., LEE, J. K., AND CHOI, S. Y. 1996 A spatial scheduling system
and its application to shipbuilding: DAS-CURVE, Expert Systems with Ap-
users to define their own work area, production schedule, and
plications, 10, 311–324.
system constraints. In addition, tying the spatial scheduling tool to L EE , J. K., L EE , K. J., P ARK , H. K., H ONG , J. S., AND L EE , J. S.
existing legacy planning and scheduling systems (e.g., SAP, Ar- 1997 Developing scheduling systems for Daewoo Shipbuilding: DAS
temis) would allow for automatic schedule updates of later- Project, European Journal of Operational Research, 97, 380–395.
scheduled blocks as schedules of upstream processes change. LIM, A., RODRIGUES, B., AND YANG, Y. 2005 3-D container packing
heuristics, Applied Intelligence, 22, 125–134.
To be successful, the automated unit scheduler must generate LODI, A., MARTELLO, S., AND VIGO, D. 2002 Heuristic algorithms for the
layouts quickly (i.e., in a matter of minutes). Methods for increas- three-dimensional bin packing problem, European Journal of Operational
ing the efficiency of the rule-based algorithm should be studied, Research, 141, 410–420.

NOVEMBER 2007 JOURNAL OF SHIP PRODUCTION 201

Das könnte Ihnen auch gefallen